> Why couldn't we at least mmap the skbuffs?
Because another pass would be still needed for checksumming. And if you
do checksumming the copy is nearly free. Zero-TCP has other costs, and
they only make sense to pay when there is appropiate hardware support
(especially hardware checksumming). ATM Linux supports only one card
with hardware checksumming support - David Miller's Happy Meal driver.
Note that there is another possible optimization that is probably
easier: Currently Linux does not do copy-checksumming in the TCP receive
path (except when it is implemented in the low level driver), but does
separate checksumming and copy passes. If this could be fixed TCP
would be probably faster (for UDP it is the same, although there
are patches around to fix that)
-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html